Late onset neonatal sepsis detection in newborn infants via multiple physiological streams

Abstracts 2002 Hockin-Mann model, with 34 states and 43 chemical kinetic equations, which was developed to model coagulation for hemorrhagic diseases and venous or coronary thrombosis. A popular indicator of coagulation activity was used to validate this model. The model's output of thrombin concentration with time, after activation by tissue factor, was compared with Calibrated Automated Thrombogram (CAT) data for severely injured patients admitted to San Francisco General Hospital. The model's use was determined by clustering and comparing its output of the concentrations of blood factors VIIIa and Va 60 seconds after model initialization to patient outcomes. The clustering methods chosen were k-means clustering and support vector machines. The second method discussed is the black-box modeling of CAT data using techniques from the field of system identification. A simplified single-input, single-output linear time-invariant dynamical system model of thrombin generation was identified and then parameterized for each patient with Simulink Design Optimization. The effects of initial concentrations of a patient's blood factors on the model's parameters were determined through forward regression. This determination facilitated a generic prediction of the concentration of thrombin over time after activation by tissue factor given any trauma patient's initial blood condition. Results: Our results indicate that the thrombin output of the Hockin-Mann model does not match trauma patient CAT data for nearly all San Francisco General Hospital patients. In addition, the Hockin-Mann model clustered outputs are no more useful than the clustered initial data when indicating mortality outcomes and clinical intervention effects. The results also show that the blackbox thrombin model output more closely predicts actual CAT data than the Hockin-Mann model, yet is much simpler at only 3 states and 5 parameters. Despite its simplicity, the black-box model parameters account for known coagulation cascade effects such as factor inhibition and the extrinsic and intrinsic pathways, providing insight into drivers of traumatic coagulopathy. Conclusions: Our work demonstrates that it is possible to model and identify important coagulopathy variables specific for patient with trauma and to track these variables accurately over time. The results of this work provide insight into the dynamic mechanisms behind coagulation after trauma, reduce the diagnostic space, and provide a step towards identifying molecular drivers of outcome. Future work will investigate the use of these mechanisms to accurately predict outcomes in patients with trauma. http://dx.doi.org/10.1016/j.jcrc.2012.10.036

Abstract 21 Late onset neonatal sepsis detection in newborn infants via multiple physiological streams Carolyn McGregor a, Christina Catley a, James Padbury b,c, Andrew James d,e a University of Ontario Institute of Technology, Oshawa, Ontario, Canada b Women and Infants Hospital, Providence, Rhode Island, USA c Brown University, Providence, Rhode Island, USA d Division of Neonatology, The Hospital for Sick Children, Toronto, Canada e Department of Paediatrics, University of Toronto, Toronto, Canada

Objectives: Early diagnosis of neonatal sepsis is challenging because the clinical features are nonspecific in the early stages of infection. Recent work suggests that patients with late-onset neonatal sepsis

e11 (LONS) are characterized by low heart rate variability (HRV) and normal respiratory rate variability (RRV), whereas patients receiving narcotics have reduced HRV and RRV [1]. Using 60-second spot readings from the Women and Infant's Hospital, Rhode Island, where data were collected remotely using Artemis Cloud [2], we use temporal abstraction to create hourly HRV and RRV summaries. We examine the relationship between low HRV and LONS, testing the hypothesis that low HRV is present before LONS diagnosis. We further investigate whether RRV improves model discrimination. Methods: We analyzed data from 128 patients admitted to the Women and Infant's Hospital, Rhode Island neonatal intensive care unit (NICU) for more than 4 days between March 2010 and April 2011. Variability was calculated by taking the absolute value of the difference between consecutive time points and creating hourly abstractions based on number of minutes where the absolute value of the difference is less than a given threshold. The population was classified by LONS status; t tests examined relationships between continuous variables. Logistic regression assessed the association between the portion of the NICU admission classified as low HRV (defined as 37 minutes low HRV per hour) and LONS, and with the proportion of the NICU admission classified as low RRV (defined as 20 minutes low RRV per hour) and LONS. The research was approved by the Hospital's Research Ethics Board. Results: Late-onset neonatal sepsis occurred in 10.3% of patients. The mean length of data collection for patients with LONS was 77.1 days as compared with 40.7 days for patients without LONS (P b .0001). The mean % low HRV for patients with LONS was 27.6% as compared with 9.8% for patients without LONS (P b .0001). Regression models showed that over the patient stay, each 1-unit increase in % low HRV was associated with a 6.2% increase in the odds of LONS (c = 0.823). After relatively aligning to the point of diagnosis, the mean % low HRV for patients with LONS was 50.0% as compared with 9.8% for patients without LONS (P b .0001). The average length of stay before LONS was 18.5 days. The mean gestational age for patients with LONS was 29.2 weeks as compared with 32.9 weeks for patients without LONS (P b .0001). Each 1-unit increase in % low HRV was associated with a 10.3% increase in the odds of LONS (c = 0.944), showing that low HRV is associated with LONS before clinical diagnosis. The results support the hypothesis that low RRV alone is not associated with LONS: the mean % low RRV for patients with LONS was 3.4% as compared with 2.5% for patients without LONS (P = .3514). Each 1-unit increase in % low RRV was associated with a 14.5% decrease in the odds of LONS; however, the results were not statistically significant. Conclusions: Low HRV was positively associated with LONS before clinical diagnosis. Future work will analyze shorter time windows and include infusion pump data to report on the results of drug existence and dosage on HRV and RRV. http://dx.doi.org/10.1016/j.jcrc.2012.10.037

References [1] McGregor C, Catley C, James A. Variability analysis with analytics applied to physiological data streams from the neonatal intensive care unit. Proc Computer Based Medical Systems, CDROM, 2012;5.

e12 [2] McGregor C, Catley C, James A, Padbury J. Next generation neonatal health informatics with Artemis. Stud Health Technol Inform 2011;169: 115-9.

Abstract 22 Tissue-specific patterns of caspase-1 and cytokines in excisional wounds are altered by shock in rat skin and muscle Ravi Starzl a,f, Dolores Wolfram b, Ruben Zamora c,d, Bahiyyah Jefferson c, Derek Barclay c, Chien Ho e, Gerald Brandacher f, Stefan Schneeberger f, W.P. Andrew Lee f, Jaime Carbonell a, Yoram Vodovotz c,d a Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA, USA b Department of Plastic and Reconstructive Surgery, Innsbruck Medical University, Innsbruck, Tyrol, Austria c Department of Surgery, University of Pittsburgh, Pittsburgh, PA, USA d Center for Inflammation and Regenerative Modeling, McGowan Institute for Regenerative Medicine, University of Pittsburgh, Pittsburgh, PA, USA e Department of Biological Sciences, Carnegie Mellon University, Pittsburgh, PA, USA f Department of Plastic and Reconstructive Surgery, Johns Hopkins University School of Medicine, Baltimore, MD, USA

Objectives: Skin and muscle wounds often lead to significant inflammation in the affected tissue. The primary mechanism by which inflammation is initiated, sustained, and terminated is cytokine-mediated immune signaling, but this can be altered by cardiogenic shock. The complexity and context sensitivity of immune signaling in general stymied a clear understanding of these signaling dynamics. We hypothesized that advanced numerical and biological function analysis methods would help elucidate the inflammatory response to skin and muscle wounds in rats, both with and without concomitant shock. Methods: We studied 2 experimental groups: wound only (“wound group”) and wound with cardiogenic shock (“shock group”). In the “wound group,” 4 Lewis rats were anesthetized and an excision biopsy was taken from the lateral aspect of the thigh on one of the hind limbs in each of the rats. In the “shock group,” 4 Lewis rats were sacrificed, and excision biopsy (wound) was carried out 15 to 30 seconds after cessation of heartbeat. Skin and muscle tissue was then separated and assayed for total protein content by BCA assay, and the inflammation biomarkers interferon-γ, interleukin (IL)-1α, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-10, IL-12p70, IL-18, monocyte chemotactic protein (MCP-1), growth-related oncogene/KC, tumor necrosis factor α, and granulocyte-macrophage colony-stimulating factor were assayed by Luminex (Luminex Corp, Austin, TX). Caspase-1, which drives activation of the NLRP-3 inflammasome, was assessed by Western blot assay. Statistical and computational analyses of cytokine network profiles (1-way balanced analysis of variance, unpaired 1-tailed heteroscedastic t test, principal components analysis, and confirmatory factor analysis) were performed in Matlab (The MathWorks, Inc, Natick, MA). Results: Granulocyte-macrophage colony-stimulating factor was elevated in the wound group (P b .05), whereas IL-4, IL-12p70, interferon-γ, and IL-18 were elevated in the shock group (P b .05). Interleukin-1α and IL-18 were more elevated in skin vs muscle (P b .05), which was suggestive of inflammasome activation in the skin. MCP-1, IL-1β, IL-2, IL-6, IL-10, IL-4, IL-12p70, and tumor necrosis factor α (P b .05) were also differentially higher in skin vs muscle. Immunoblotting revealed caspase-1 activation in skin but

Abstracts not muscle. Notably, IL-1α and IL-18, along with caspase-1, were greatly elevated in the skin after cardiogenic shock (P b .05). Principal components analysis suggested that more than 95% of observed variance could be explained by 2 principal components in the skin and muscle and suggested distinct groups of inflammatory mediators induced in the “wound group” vs the “shock group”. Interleukin-18 and IL-1α were primary contributors to the first principal component, whereas IL-6, MCP-1, growth-related oncogene/KC, and IL-1β were primary contributors to the second principal component. Principal components analysis results were reinforced by confirmatory factor analysis. Conclusions: Caspase-1 and the NLRP-3 inflammasome appear to be key factors in determining the type and severity of the inflammatory response to excisional wounding, especially in the presence of shock. Activated caspase-1 is associated with defined, compartmentalized patterns of cytokine production that may be discerned via data-driven modeling. http://dx.doi.org/10.1016/j.jcrc.2012.10.038

Abstract 23 Global sensitivity analysis of endotoxin-induced acute inflammatory responses predicts the multimodal dependence of global tissue damage both on host IL-6 responses and endotoxin dose Shibin Mathew a, John Bartels c, Ipsita Banerjee a,b, Yoram Vodovotz d,e a Department of Chemical and Petroleum Engineering, University of Pittsburgh, Pittsburgh, PA, USA b Department of Bioengineering, University of Pittsburgh, Pittsburgh, PA, USA c Immunetrics, Inc, Pittsburgh, PA, USA d Department of Surgery, University of Pittsburgh, Pittsburgh, PA, USA e Center for Inflammation and Regenerative Modeling, University of Pittsburgh, Pittsburgh, PA, USA

Introduction: Acute inflammatory responses to various stimuli involve complex nonlinear interactions among inflammatory cells and their products. We previously developed a nonlinear ordinary differential equation model to explain the dynamics of endotoxin (lipopolysaccharide, or LPS)-induced acute inflammation and associated whole-animal damage/dysfunction (a proxy for the health of the organism, whose likely molecular correlate is damageassociated molecular pattern molecules) [1]. That model includes the inflammatory mediators tumor necrosis factor α, interleukin (IL)-6, IL-10, and nitric oxide (NO) and was calibrated in part on LPS doses of 3, 6, and 12 mg/kg in C57Bl/6 mice. In the current work, we analyzed the major determinants of the resulting tissue damage using a global sensitivity approach. Methods: The precise inflammatory role of IL-6 and its use as a biomarker or therapeutic target have been the source of much debate, presumably due to the complex proinflammatory and anti-inflammatory effects of this cytokine. Therefore, we chose to investigate the sensitivity of the area under the IL-6 curve (AUCIL6) and the area under the damage curve (AUCD) to the 51 rate parameters of the ordinary differential equation model for different levels of simulated LPS challenge (1-15 mg/kg). Owing to the complex nonlinear interactions between the rate parameters, we chose a variance-based global sensitivity approach. Specifically, we chose the Random Sampling High Dimensional Model representation developed by Rabitz et al to reduce the computational cost associated with Monte Carlo sampling of the parameter space [2]. Monte Carlo samples